Stream Finance: loss broke three stablecoins

Stream Finance issued xUSD as a yield-bearing synthetic dollar. Deposits funded positions run by an external fund manager, and the returns were packaged into an on-chain token that could be lent, borrowed against, and re-deposited across DeFi. The hedge and the trading book lived off-chain, behind a single manager, with no multisig over the capital, no published custody arrangement, and no live attestation tying circulating xUSD to verifiable backing.

On 10 October 2025 a sharp ETH decline liquidated that off-chain book. The loss was real immediately but invisible on-chain for weeks, because the lending markets holding xUSD priced it at a fixed $1. On 4 November Stream disclosed roughly $93M of losses, halted withdrawals, and opened an investigation; xUSD fell about 77% toward $0.26 and stranded on the order of $160M of deposits. A separate Balancer exploit the same week added noise, but the trigger was the trading loss, not a smart-contract hack.

The failure did not stay inside Stream. Because xUSD had been re-hypothecated as collateral across protocols, the hole propagated through the collateral graph: Elixir's deUSD, roughly 65% backed by loans to Stream, collapsed near 98% to about $0.015 and was wound down, and Stables Labs' USDX broke below $0.60 within days. Analysts mapped roughly $285M of interconnected debt; lenders froze markets and roughly $1B left DeFi yield products in a week.

The mechanism that turned one fund's loss into a market-wide event was leverage built on top of leverage. xUSD was not simply deposited and held; on Morpho, Euler, and Silo, curators created isolated vaults where users supplied real stablecoins — USDC, USDT, USD1 — against xUSD collateral, and the borrowed stablecoins were looped back into more xUSD positions. Each turn of the loop manufactured more apparent xUSD footprint without adding any real backing behind it.

Elixir sat one layer further out. Its deUSD was a synthetic dollar in its own right, but roughly 65% of deUSD's backing — about $68M — had been lent to Stream through private Morpho vaults. So deUSD was, in effect, a claim on Stream's solvency wearing a different ticker. When Stream's backing evaporated, deUSD's reserves evaporated with it, and Stream reportedly held around 90% of the remaining deUSD supply, leaving it unable to settle. USDX, sharing collateral and venues with the same complex, formed a third node on the same web.

This is the structural hazard composability carries when collateral is recursive and opaque. A token that is sound in isolation can become a transmission line: pledged into one market, borrowed against, re-pledged into the next, until a single off-chain failure radiates through every protocol that accepted it. The dependency was always visible in the collateral graph; what was missing was anyone pricing it as a dependency rather than as a dollar.

A lending market's safety rests on one mechanism: when collateral falls below a threshold, it is liquidated and the loan is made whole before the position goes underwater. That mechanism only works if the market can see the collateral's real price. The vaults holding xUSD did not — many priced it at a hardcoded $1, a fixed peg assumption rather than a live feed reflecting traded value.

Hardcoding $1 is sometimes defended as protection against transient depegs and liquidation cascades on assets expected to mean-revert. Here it did the opposite. As xUSD's market price collapsed, the protocols still valued it at par, so no liquidation ever triggered. Borrowers, seeing collateral the protocol would not mark down, drained the supplied USDC against it; lenders were left holding bad debt denominated in a token worth a fraction of the assumed price. The oracle had not protected the market — it had disabled its only defense and converted a depeg into a withdrawal of real assets.

Pharos treats this as a downstream-pricing risk distinct from the issuer's own mechanism. A coin can fail at the issuer while integrators that price it at a static $1 turn that failure into systemic bad debt before the depeg is reflected in their books. The fragility is not in the stablecoin alone; it is in every market that agreed to pretend it could never move.

The contagion ran along the curator model. Permissionless lending venues like Morpho and Euler let third-party risk managers spin up isolated vaults and allocate depositor capital to high-yield strategies. That allocation decision — which collateral to accept, at what loan-to-value, against which price feed — is where the risk was concentrated, and several curators had routed large sums into Stream-linked exposure. TelosC carried the single largest position near $124M; MEV Capital and Re7 Labs held tens of millions each.

When xUSD broke, the curated vaults could not contain it. Compound suspended several stablecoin markets, Euler froze pools carrying tens of millions in bad debt, and the freeze spread to any venue holding the linked collateral. Roughly $1B of capital left DeFi yield products within a week — not all of it directly exposed, but withdrawing on the recognition that exposure was hard to see and harder to exit. The episode drew comparisons to Terra, and prominent builders warned publicly that the curator economy was carrying Terra-shaped risk.

The lesson generalizes past these three coins. A curator earning yield by accepting an opaque, re-hypothecated token as collateral is underwriting a counterparty most depositors never see and cannot assess. When that counterparty is a single off-chain manager, the vault's headline APY is compensation for a tail risk the depositor was never shown — and composability ensures the tail, when it arrives, is not contained to one vault.

Composability is a risk multiplier, not just a feature. A token that is re-hypothecated as collateral across protocols becomes a transmission line: one issuer's failure radiates through every market that accepted it, and recursive loops inflate the footprint far beyond the real backing. Assess a synthetic dollar not only on its own mechanism but on how widely and how deeply it has been pledged, looped, and re-pledged elsewhere.

A hardcoded $1 oracle disarms a lending market. Pricing collateral at a fixed peg instead of a live feed does not prevent a depeg — it prevents liquidation during one, letting borrowers drain real assets while lenders accumulate bad debt in a worthless token. Static pricing of a volatile-backed asset is a maximum-severity integration flaw regardless of how stable the asset has looked.

Off-chain backing without on-chain proof is unverifiable trust. xUSD's yield came from a single manager's book with no multisig, no disclosed custody, and no live attestation; the loss existed for weeks before anyone on-chain could see it. When the engine of a yield-bearing dollar is off-chain and opaque, the headline rate is payment for an unobservable counterparty risk, and the curators and integrators accepting it are passing that risk to depositors who were never shown it.